Good afternoon! Today we're talking near-term innovation, so we asked the experts to identify a process or task that humans still did better (or at least still did) that AI was going to do better soon. Questions or comments? Send us a note at firstname.lastname@example.org.
CTO at Scale AI
The space of generative AI is incredibly promising. Art, music, video and entertainment are well within the realm of generative AI already. But "better" is subjective. I believe we will increasingly see human artists guiding and collaborating with generative AI to continue to push the world of art further, faster. That, combined with the emerging marketplaces for NFTs, could lead to a new Renaissance in the art world.
One of IBM's ethical principles of AI is that this technology should be used to augment human intelligence, not replace it, affording employees more time to focus on higher-value work. In a business setting, that means that AI needs a certain level of reasoning ability which, today, it frankly doesn't have. But a newer, more hybrid approach to AI is beginning to bridge that gap.
Neuro-Symbolic Learning (NSL) combines the strengths of neural networks and classical symbolic AI to accomplish complex tasks that neither of these methods can perform individually. At IBM, we are working with NSL to demonstrate AI's ability to solve much harder problems, learn with dramatically less data, and provide inherently understandable and controllable decisions and actions.
That's why I think the next "domino to fall" will be AI's ability to understand and answer complex questions with minimal domain-specific training. For instance, if I asked the question, "What were our sales figures last quarter," I don't need a report, I need a number. However, if I asked, "What were our sales last quarter by geography compared to our projections," this requires at least a table if not several visualizations. Mapping the language to the data elements is relatively easy today; however, a hybrid approach to AI like NSL is needed to inject reason into producing more complex and actionable results.
With more "common sense," AI will be better at uses like customer support, business intelligence, advanced discovery and much more.
Executive Director, Deloitte AI Institute at Deloitte
Artificial intelligence is on the verge of revolutionizing the world of fashion, giving retailers the ability to instantly determine which clothing items are the best fit for a customer's particular size and body shape.
Finding items that fit is one of the most frustrating parts of clothes shopping. From a consumer's perspective, it can be a time-consuming hassle during the purchase phase — and all too often leads to dissatisfaction and return hassles as well.
From a retailer's perspective, it can arguably be an even bigger problem, requiring large inventories of sizes and styles; sales clerks with sufficient experience and expertise to steer customers toward the right items; unhappy customers frustrated with the fitting process; and the time and expense of dealing with returns.
Systems that incorporate machine learning, computer vision and 3D scanning help minimize these challenges by obtaining a shopper's measurements in real time simply by having them stand in front of a camera. Those measurements can then be matched against a database of clothing to find the best fit, improving customer satisfaction and reducing the cost of returns.
Early adopters — such as online personal styling brand Stitch Fix — are already embracing a hybrid human/AI stylist model to personalize clothing and fit for their customers. We expect more retailers to follow as the benefits and impact become clear. There is no doubt that brands and retailers that embrace AI, machine learning and related technologies will have a competitive edge in the fast-moving world of fashion.
With the acceleration of data warehousing, and as it becomes easier to integrate more back-end systems, we'll start to see more businesses leverage AI as a tool to identify patterns and trends within aggregate data. By layering different data sources together, whether across marketing, customer or financial data, you have the ability to uncover a ton of hidden insights, but you need technology that can quickly check for patterns. AI doesn't do anything that a person can't do: It's simply executing tasks faster and on a much larger scale.
For example, Dialpad recently acquired a company called Koopid that orchestrates written conversations across all digital channels. Because Dialpad captures conversational data in real time from calls and meetings, we have the opportunity to connect two disparate data sources that would not typically feed into each other. So, if a customer calls your contact center with a question, we can pull all the past touchpoints that the customer has had with your business regardless of channel (voice/text/email/etc.). Have they complained on social? What was the purpose of the past five calls they made to you?
Once you have a holistic view of a customer's aggregate data, you can feed that data to different parts of the business. This helps service agents provide better support and give management a view into which customers are more likely to churn; meanwhile, routing the data in real time to the product team could help resolve issues faster. Essentially, the possibilities are limitless once we start using AI to aggregate and understand large data sets.
We have studied the property and casualty insurance industry for some time and are convinced that the sector is on the cusp of digital and AI transformation. Insurance is largely a data game that is played by estimating risk, predicting correlation and spotting trends, so it is an ideal fit for AI's current core competencies.
In today's insurance markets, there is a tremendous amount of data available, but it is fairly unorganized and hence generally not used. Amazingly, much of the work related to collecting, structuring and understanding this data is still carried out by human-led processes, which is not only highly inefficient, but also suboptimal in terms of the analysis.
We believe that AI will help lower the cost of data access and ignite significant change. Ultimately, companies will better identify, manage and avoid risks by understanding insights from analyzing this data. This will allow the insurance industry to play a critical strategic role as the aggregator of data, helping clients to better understand their own risks and offering solutions to more effectively manage them.
Data science alone is not enough to understand the world and the objective here is not "robotic underwriters" but "robotic arms for human underwriters." Humans are mandatory throughout the entire AI development cycle to help avoid biases, ensure explainability and drive governance. In insurance (and all other sectors) we must not underestimate the importance of "uniquely human skills" — the social, creative and intangible skills that cannot be automated away — and realize that, for the foreseeable future, AI and sophisticated robotic systems will require a human counterpart to achieve their full potential.
In recent years, the capabilities of AI have grown by leaps and bounds. And this is especially true when it comes to the field of HR, including using AI capabilities for talent acquisition and its subsequent management once individuals are in the door. However, it's not a hands-off approach just yet. In leveraging AI for talent intelligence platforms, companies can identify the existing skills and capabilities in their organization, analyze what is required, and highlight the best candidates — with an eye on culture, diversity and inclusion. It can even look to industry trends and recognize what skills will be necessary in five years, and the experiences your teams need to acquire them.
So what can't it do here? Pound the table. Champion the cause, and put the discussion front and center. Especially now, as the workforce ramps back up and many individuals look for new careers, future-facing organizations have an opportunity to identify and acquire cornerstone talent, building blocks that will set themselves apart from the crowd. The ability to do so is already here, but there are still many false narratives around artificial intelligence: outdated stories of bias, discrimination, etc. It is the responsibility of organizational leadership to look past some of this clutter and realize that AI brings unprecedented short and long-term value by making sure that each employee is a great fit and on an upward trajectory within it. Making the decision to use this technology and get past gut feelings, educational biases and more — that's the move that is going to set leading organizations apart from the rest going forward.
The rote and repetitive parts of quantitative analysis and presentation are ripe for improvement: joining data sets, slicing and dicing, and generating customized dashboards and reports are dominos that are all ready to fall. These tasks aren't the satisfying part of quantitative analysis, but they suck up most of people's time because today's tools are overwhelmingly manual and each successive question requires doing a huge amount of work from scratch. Given the increasing degree of data centralization in modern organizations as well as the rapid development of generative models, I expect we'll see a new class of models to improve on the end-to-end turnaround time and quality of these kinds of analysis results.
Kevin McAllister (
@k__mcallister) is an associate editor at Protocol, leading the development of Braintrust. Prior to joining the team, he was a rankings data reporter at The Wall Street Journal, where he oversaw structured data projects for the Journal's strategy team.